On the Origin of Shared Beliefs (and Corporate Culture)
Preliminary
Eric Van den Steen∗
September 7, 2003
Abstract
Since organizational homogeneity may reduce or eliminate agency problems, understanding
its sources is important for the trade-off between firms and markets and for organization design. This paper shows that a firm’s members will develop homogenous beliefs through shared
experiences. Organizations with a long and successful history and high employee involvement
will have the most homogenous beliefs.
I relate this also to the notion of corporate culture as shared assumptions and beliefs. The
model confirms a pattern of ‘facts’ that has been informally suggested by the management
literature. Similar organizations may develop very different cultures, which can persist even
when all the original members are gone and even when the culture is almost surely suboptimal.
The organization’s culture is heavily influenced by the founder’s beliefs and by early experiences,
and is more likely to change under an outsider-successor to the CEO than under an insider. The
paper also studies the impact of the rate of learning and questions the received interpretation
of the correlation between culture and performance.
1
Introduction
Agency theory has been wonderfully successful. It takes, however, the differences in beliefs and
objectives that are at its core, as exogenously given.1 This is an important issue in view of the fact
that organizations seem to be more homogenous in beliefs and values than society at large. Such
homogeneity may make incentives more effective and may simplify delegation, communication, and
coordination (Crawford and Sobel 1982, Baker 1992, Crémer 1993, Aghion and Tirole 1997, Dessein
2001). Homogeneity reduces, however, the incentives to collect information (Van den Steen 2002b),
which may hinder learning and innovation. Understanding the sources of homogeneity is thus
∗
MIT-Sloan School of Management ([email protected]). This paper benefitted greatly from two extensive discussions
with Bob Gibbons, a challenge by George Baker, the guidance of Bengt Holmström and John Roberts, the discussion
by Ben Hermalin, and very useful remarks from Oliver Hart, Ed Lazear, Paul Milgrom, John Matsusaka, Kevin
Murphy, Sven Rady, Jesper Sørensen, Tom Stoker, Birger Wernerfelt, the participants in the NBER organizational
economics conference, the MIT organizational economics lunch, and the seminars at HBS and USC.
1
Agency problems at the managerial level, which are the focus of this paper, are typically about ‘decisions’,
i.e. about the optimal course of action, rather than about ‘effort’. As such, these agency problems are often driven by
differing beliefs and values regarding alternative actions. While most of the models in agency theory are formulated
in terms of differing utility functions, the can often be interpreted as reduced forms for differing beliefs.
1
important for the trade-off between markets and organizations and for the design of organizations
and incentive systems.
The key argument of this paper is that this homogeneity is at least in part caused by the
fact that the members of an organization have a natural tendency to develop homogenous beliefs
through shared experiences.2 Such homogeneity will be especially pronounced in organizations with
a long and successful history and with high employee involvement in decision making. Since shared
experiences depend on long-term interactions, it is ultimately the permanency of the organization
which is the source of this organizational homogeneity.
After establishing these basic results on ‘shared beliefs’, I link them to the notion of corporate
culture. Seminal authors such as Schein (1985) or Kotter and Heskett (1992) have argued that
shared beliefs and assumptions are an essential part, if not the essence, of corporate culture. The
model in this paper formalizes and formally explains a pattern of ‘facts’ or informal findings that
has been suggested in this literature.
• Organizations in identical circumstances may end up with very different cultures.
• Culture may persist, even though all the original members have left.
• Culture may persist, even though it is almost surely suboptimal.
• The beliefs of the original leader and its early experiences are important determinants of an
organization’s culture.
• External succession of the CEO is more likely to lead to a change in culture than internal
succession.
I also suggest some new results in this context. I show, for example, that in environments with a
slow learning rate there will be more diverse cultures across firms and the influence of the CEO’s
original beliefs will be stronger.
The results have further implications. The stability of an organization’s values and beliefs might
help explain the persistent differences in performance of firms in the same industry (Mueller 1990,
McGahan 1999). The results on succession and change of culture imply that the boards of well
performing firms should be more likely to select an insider successor to a CEO than boards of worse
performing firms. Finally, the earlier mentioned result that successful organizations will generate
strong cultures raises doubts about the received interpretations of the correlation between culture
and performance.
These results are developed in the context of a simple model that tries to capture the learning
process of a group of people faced with a new task. One may think of a start-up that tries to organize
itself. The group is faced with a set of alternative courses of action, or ways of doing things, among
which it must choose. At the start, the performance of the alternatives are unknown, and the
agents may openly disagree, i.e. I assume that agents may have differing priors, an assumption that
I discuss in more detail in appendix B. By trying out different actions, the group learns over time
which ones work best. A relative consensus develops on the best way of doing things. It becomes
‘the way we do things around here’. Formally, the model is essentially that of a multi-armed bandit
with instant revelation.
2
In a parallel paper, Van den Steen (2002a), I consider sorting as an alternative mechanism.
2
The Literature. Research on the origin of homogenous beliefs is, to my knowledge, new to the
economic literature. A parallel paper (Van den Steen 2002a) deals with sorting as an alternative
source of such homogeneity.3
Congruity of objectives itself, on the other hand, has been implicit or explicit in a number of
key papers in the agency literature (Crawford and Sobel 1982, Baker 1992, Crémer 1993, Aghion
and Tirole 1997, Dessein 2001). Van den Steen (2002b) interprets some of these in the context of
differing beliefs and introduces new aspects, in particular the fact that heterogenous beliefs create
incentives to collect information, which may be an important factor in innovation and change.
The economic literature on corporate culture is in its early development. The papers most
related to this one are Crémer (1993) and Lazear (1995). Crémer (1993) defines culture, following
Schein (1985), as a stock of shared knowledge and argues that it improves the efficiency of information processing. The paper, which contains a very interesting model on the coordination effects of
shared knowledge, focuses on the effects rather than the causes of shared knowledge. It essentially
starts from the premise that culture is useful, apart from some unavoidable side-effects, and tries
to explain why. Lazear (1995) also defines culture as ‘shared beliefs, values and technology’ and
considers a ‘genetic’ evolutionary model of corporate culture, built on the assumption that culture
is contagious. His work is complementary to the current paper in that it provides an alternative
perspective how culture might evolve. The difference is clearly the level of abstraction, where
Lazear takes the genetic nature of culture as a given. He also assumes that culture is useful.
An important alternative model of culture is that presented by Kreps (1990) and further clarified
and interpreted by Hermalin (2001). Hermalin points out that there are in fact two notions of
corporate culture in Kreps’ paper: culture as a mechanism to coordinate in the presence of multiple
equilibria and culture as a reputation for dealing in a specific way with unforseen contingencies.
Culture as a coordination mechanism is a pure convention, like driving on the left, and thus very
fragile.4 Culture as reputation is valuable, argues Kreps, since it protects employees against abuses
of authority in unforseen contingencies. Hermalin argues, however, that this latter notion is based
on a fair amount of ‘hand-waving.’ The key difference with the current paper is that the KrepsHermalin model focuses on the effects rather than on the origin of culture, and again seems to start
from the premise that culture is essentially good, apart from unavoidable side-effect. Aside from
his important discussion of Kreps’ model, Hermalin (2001) also summarizes and reinterprets other
existing research, and adds to it by linking the topic with insights in other fields of economics,
such as IO. Along such lines, Carrillo and Gromb (1999) model corporate culture as production
technologies for which employees can make specific investments. The fact that employees choose
their investments simultaneously combined with the possibility for the firm to change technology
can lead to the coexistence of a strong culture (high investment) equilibrium and a weak culture
(low investment) equilibrium. Rob and Zemsky (2002) present a theory in which firms differ in the
stationary levels of cooperation among their employees, which they equate with corporate culture,
in the sense of a ‘stable, [. . . ], pattern of behavior’. This notion of corporate culture is different from
the one used in this paper, but is also very interesting. The model in this paper is further related
to the work of Prescott and Visscher (1980) and could be extended to overlap with organization
capital in their sense. Note, finally, that Greif (1994) considers ‘cultural beliefs’ an important part
of a national culture.
Weber and Camerer (2001) present experimental results that bear on the phenomenon culture.
3
Related is also Van den Steen (2002d), which shows how a manager’s strong beliefs or vision may lead to sorting
in the labor market and thus to homogenous beliefs.
4
Including ‘driving on the left’ as a part of culture opens the possibility that culture might be changed by law.
3
They let pairs of people (‘firms’) develop, through trial and error, a homemade language for solving
problems, which they interpret as the firm’s culture. They then merge groups and show that their
performance declines after the merger. A key difference with the current theory is that culture in
their sense has an obvious performance advantage.
Apart from drawing attention to this important issue of homogeneity, this paper’s contributions
are to show that organizations have a natural tendency to develop homogeneity through shared
experiences, to identify circumstances that enhance this homogeneity, to link it to the notion of
corporate culture, to explain the pattern of ‘facts’ on corporate culture that have been informally
suggested by the managerial literature, and to generate some new predictions in this context.
Note also that this paper does not assume that homogeneity or culture is good. It just says that
homogeneity and culture will tend to develop naturally and that this has important implications.
The next section describes the model of the paper. Section 3 develops the basic homogeneity
results, while section 4 considers the pattern of results related to corporate culture. Section 5
concisely discusses potential tests, while section 6 concludes. The appendices contain a further
discussion of culture and its definitions and a discussion of the use of differing priors.
2
The Model
The model tries to capture the situation of a recently formed group that is faced with a new task.
One could think of a new product development group which slowly finds its ways, as described in
McCaskey (1997), or a start-up, as in Schein (1985).
The group will try alternative ways of going about their work, starting with those that its
manager considers most effective. Depending on the outcomes, the group will learn and may try
other alternatives. Formally, the group is thus faced with a repeated choice among alternative ways
of getting things done. Over time, the members of the group come to share beliefs about what
works and what doesn’t.
Since I want to contrast within-organization belief differences and across-organization belief
differences, consider multiple organizations. Each organization consists of one manager, denoted
m, and J members, denoted 1 through J. Each organization is faced with an infinitely repeated
choice among N action an ∈ A, that is common to all firms.
The manager chooses in each period t the action at ∈ A that her organization will undertake
in that period. The objective of the manager is to maximize her δ-discounted payoff. Action
a has a payoff ρa . These payoffs are initially unknown to the agents, but each agent has his or
her own subjective prior beliefs.5 Agent i’s prior belief is that ρa is distributed according to
some distribution Ga,i (ρ) with mean ra,i . Assume that the Ga,i are identical up to their mean,
i.e. Ga,i (u) = G(ra,i , u) where G(x, ·) is a distribution function with mean x and density g(x, ·).
Since only the means distinguish these distributions, I will refer to these means ra,i as ‘beliefs’.
While agents do not observe each others’ beliefs, it is commonly known that they may differ. This
implies that I do not impose the common prior assumption, an approach discussed in appendix B.6
The payoffs ρa and the means of the priors ra,i are given at the beginning of the game. I assume
that, empirically, ρa and ra,i are i.i.d. distributed according to some distribution F . Remember
that this latter distribution is not a prior, but an empirical distribution that just happens to reflect
5
We could also allow some of the payoffs to be commonly known, without changing the results.
It is commonly known that agents have no private information beyond what they observe about the actions of
their organization. An agent will not be able to infer anything from the fact that other agents in his organization
hold beliefs that are different from his own.
6
4
the distribution in the population. I will give below two slightly more concrete examples of this
setup.
I also assume that the organization’s actions and payoffs are costlessly observed by all its
members, but by no one else. The assumption that outsiders cannot observe the focal organization’s
actions and performance is critical to the analysis, but should not be controversial. Competitors
of Caterpillar, Walmart, or McKinsey, for example, have tried for years to understand these firms’
recipes for success, with little success. There are multiple reasons why it is so difficult to figure
out what other organizations are doing. One reason is that it is very difficult to describe and
communicate something as complex as an organization’s way of doing things. It is like trying
to describe a Picasso. Such things are nearly impossible without direct observation and active
participation, or (in the case of organizations rather than Picasso’s) years of coaching and detailed
stories. Another reason is that firms have incentives to keep their successful practices secret. Note
that I do not assume that no actions or outcomes can be observed. But I abstract from the
dimensions that are easy to observe.
I will discuss later two extensions of this model that essentially relax the assumptions that only
the manager can undertake actions and that all the members of the organization costlessly observe
all its outcomes.
Two examples The following are two slightly more concrete examples of the above situation. In
each case, the organization consists of one manager and two employees and there are two possible
actions, a1 and a2 .
1. The outcome of an action is a real-valued payoff ρa , which is unknown. Agent i thinks that
ρa is distributed ρa ∼ N (ra,i , 1). So agent i believes that on average the payoff of a will be
ra,i , but isn’t completely sure about it, and the standard normal expresses that uncertainty.
Let F be a standard normal distribution. This means that the two ra1 ,i and ra2 ,i of each
of the agents are drawn from a standard normal distribution. Note that ra1 ,i and ra1 ,j will
typically differ. The payoffs ρa are also drawn from the standard normal distribution.
2. The outcome of an action is either a success or a failure. The probability of success of an
action is ρa = pa , but is unknown. Each agent thinks a priori that pa is distributed according
to a Beta distribution with mean ra,i ∈ [ 0, 1 ], and parameters a = 10ra,i and b = 10(1 − ra,i ).
This means that agent i thinks that on average action a has the likelihood of success equal to
ra,i , but is not completely certain, and the Beta-distribution expresses that uncertainty. Upon
trying an action, all members of the organization perfectly observe pa . (So they do not just
observe one success or failure, but the probability itself. We could imagine that they observe
the consequences of the action and can derive pa from that.) Let, finally, F be uniform. This
means that the ra,i are independent draws from U [ 0, 1 ], and the ρa likewise.
Some notation and preliminary analysis This type of problem is known as the multi-armed
bandit problem. In particular, the model corresponds to a multi-armed bandit with N independent
arms. In such model, taking an action has two effects: a direct contribution to profits and information about the payoff of that particular action. It is thus not always optimal to simply choose
the action with the highest expected payoff: the longer term gain from learning about some other
action might outweigh the temporarily lower performance.
t denote agent i’s belief at the start of period t about ρ . Superscripts will always refer
Let ra,i
a
to the period. Let Bit denote the set of actions that have been tried by agent i’s manager by the
5
start of period t. I will drop the subscript i and just use B t when it is clear which organization is
meant. Let the sample point ω denote a realization of (ρa )a∈A and (ra,i )a∈A for all agents. The
following describes the optimal strategy in the basic model.
t , then the
Lemma 1
• Consider period t. Let â = argmaxa∈B t ρa and ã = argmaxa∈A\B t ra,m
manager’s optimal strategy is to undertake ã if
Z ∞
δ
rã,m ≥ ρâ −
(u − ρâ )g(rã,m , u) du
(1)
(1 − δ) ρâ
and â otherwise.
• There exists almost surely a period t ≤ N such that the manager undertakes the same action
a∗ (ω) forever after.
Proof : The first part follows from the fact that the Gittins index (Gittins and Jones 1974) applies, since
this is a multi-armed bandit with independent arms and geometric discounting. The Gittins index for rã is
the ρ that makes equation (1) hold with equality. Since the index increases in the mean, it is sufficient to
consider only ã and â. The first part then follows.
For the second part, consider the measure-1 set on which the prior beliefs of the agents are all distinct. Note
that, if it is ever optimal for the manager to choose a known action â, then that will be optimal forever after.
If an unknown action gets tried after period N , then a known action must have been tried before period N
(since there are only N unknown actions), which leads to a contradiction. It follows that after period N only
known actions get used, and then it is optimal to use â forever.
¥
The manager will thus always settle on an action but, as I will show later, not necessarily the best
action (Rothschild 1974).
3
Shared Beliefs
The central result of this paper is that members of an organization will come to share certain beliefs
through shared experiences and that such homogeneity will be stronger in older firms, firms that
have been successful, and firms in which employees are involved in decision making.
Before getting to these results, I need to be more precise on the meaning and measurement of
belief homogeneity, since it plays such a central role in the analysis.
3.1
Measuring homogeneity
While homogeneity of beliefs is a straightforward idea, there are multiple ways to formally define
and measure it. Each of these makes sense in a particular context. In fact, any specific agency
model will typically lead to a slightly different measure of homogeneity.
A very attractive measure of homogeneity is the likelihood that two randomly selected members
of the organization agree on the optimal action. Consider the start of period t. Let ãti denote the
action that has the highest payoff according to agent i. The first measure of homogeneity is then
defined as
PJ Pi−1
i=1
j=1 Iãti =ãtj
H1t =
J(J − 1)/2
6
where I is the indicator function. A nice aspect of this measure is that it is directly related to the
probability that two randomly selected members will ‘do the same thing’ and thus to ‘the way we
do things around here.’
A shortcoming of H1t is that it does not reflect ‘how much’ two agents agree or disagree. A
measure that is very good in this respect is the average squared Euclidean distance (at the start
of the period) between the means of the beliefs of two randomly selected members. Let d(rti , rtj ) =
PN
t
t
2
n=1 (ran ,i − ran ,j ) , then
J
H2t =
J
XX
1
d(rti , rtj )
J(J − 1)
i=1 j=1
This implicitly assumes that differences in beliefs about different actions are equally important.
In some cases the beliefs about the current action are relatively more important, which leads
to a weighted Euclidean distance. Let ǎ denote the action that was chosen P
last period and let
t
t
2
ˇ t , rt ) = γ1 (rt − rt )2 + γ2
γ1 , γ2 ∈ [ 0, 1 ] parametrize the weights.7 Define d(r
i j
ǎ,i
ǎ,j
a∈A\ǎ (ra,i − ra,j )
and
J
H3t
J
XX
1
˜ t , rt )
=
d(r
i j
J(J − 1)
i=1 j=1
Other definitions and measures are possible. I will concentrate on the measures above since they
span a large spectrum.
3.2
Basic homogeneity and its evolution over time
I now come to the most basic result of the paper: that members of the same organization will hold
more similar beliefs than society at large. The reason is that they share similar experiences. Since
the model was set up to capture this effect, the result should be no surprise to the reader.
To obtain the result formally, I consider three agents of which two, denoted 1 and 2, belong to
one organization, say f , while the third agent, denoted 3, belongs to a different organization g. Let
ãti denote the action that has the highest payoff according to agent i at the start of period t. The
proposition considers all three measures of homogeneity defined earlier earlier.
Proposition 1 For any period t, any two agents 1 and 2 of some organization f , and any agent 3 of
˜ t , rt )] ≤
some other organization g, P [ãt1 = ãt2 ] ≥ P [ãt1 = ãt3 ], E[d(rt1 , rt2 )] ≤ E[d(rt1 , rt3 )], and E[d(r
1 2
t
t
˜
E[d(r1 , r3 )].
Proof : For the first part of the proposition, fix a set of payoffs (ρa )a∈A . Fix an action, say ǎ, and let
the sample point ω denote a realization of (ρa )a∈A and (ra,i )a∈A for all relevant agents i in the model.
t
t
Let Xǎ,1 = {ω : rǎ,1
≥ ra,1
∀a ∈ A} denote the event that member 1 considers ǎ to be the best action.
It is sufficient to show that P [Xǎ,1 ∩ Xǎ,2 ] ≥ P [Xǎ,1 ∩ Xǎ,3 ]. Conditional on (ρa )a∈A , Xǎ,1 and Xǎ,3 are
independent. Moreover, by symmetry P [Xǎ,1 ] = P [Xǎ,2 ] = P [Xǎ,3 ], so that it is sufficient to show that
P [Xǎ,1 ∩ Xǎ,2 ] ≥ P [Xǎ,1 ]2 or P [Xǎ,1 | Xǎ,2 ] ≥ P [Xǎ,1 ], where 1 and 2 are in the same firm, or that
t
t
t
t
t
t
≥ ra,1
∀a ∈ A | rǎ,2
≥ ra,2
∀a ∈ A] ≥ P [rǎ,1
≥ ra,1
∀a ∈ A | ∃a ∈
P [Xǎ,1 | Xǎ,2 ] ≥ P [Xǎ,1 | Xǎ,2 ] or P [rǎ,1
t
t
A : rǎ,2 < ra,2 ] which is straightforward.
7
We could also let ǎ be the action that, according to the manager’s beliefs at the start of period t, has the highest
expected payoff. All results would go through.
7
P
t
t
For the second part of the proposition, note that for any agents i and j E[d(rti , rtj )] = a∈A E[(ra,i
−ra,j
)2 ] =
P
P
2
t
t
t
t
t
t
a∈Bit ∩Bjt 0 +
a∈A\(Bit ∩Bjt ) 2σF . Since {a ∈ Bi ∩ Bj } ⊂ {a ∈ Bi } and since Bi ∩ Bj = Bi when i and j
are part of the same organization, the second part follows. The argument for d˜ is analogous.
¥
This formally establishes that joint experience will lead to shared beliefs.
Given its potential importance in mediating agency conflicts, understanding the factors that
may stimulate such homogeneity is important for the design of organizations and incentives, among
other things. The following sections study these factors.
3.3
Evolution over time
Since shared experience drives homogeneity, it is reasonable to conjecture that organizations with
more experience will be more homogenous. The following proposition says indeed that, in this static
model without turnover, older organizations will be more homogenous. More precisely, it says that
the probability of agreement between two agents in the same organization increases monotonically
over time while the distance between their beliefs decreases monotonically over time.
Proposition 2 For any period t and any members 1 and 2 of the same organization, E[d(rt1 , rt2 )] ≥
t+1
t+1 t+1
t
t
˜ t t
˜ t+1 t+1
E[d(rt+1
1 , r2 )], E[d(r1 , r2 )] ≥ E[d(r1 , r2 )], and P [ã1 = ã2 ] ≤ P [ã1 , ã2 ].
Proof : For the first two parts of the proposition it is sufficient, by the proof of proposition 1, that the
number of action tried by organization f increases over time. This is of course the case.
Consider then the last part of the proposition, that the probability of agreement increases over time. Condition on (ρa )a∈A , on B t (the revealed set at the beginning of the period), and on {ra,m }a∈A (the set of
priors of the principal). Let the current best known action in period t be â, with performance ρâ . Let
k = #(A \ B t ) denote the number of unknown actions.
We need to prove that the probability of two randomly selected agents agreeing increases when the set of
known actions goes from B t to B t ∪ ǎ, or the number of unknown actions goes down from k to k − 1. Since
the proposition is trivial when k = 1, I will assume k ≥ 2.
The probability that two agents agree on â = argmaxa∈B t ρa as the best action is F (ρâ )2k . The probability
that they agree on some particular unknown action is (by independence of their beliefs)
R ∞ the product of the
probabilities that one member thinks that action is best. That latter probability is ρâ F (u)k−1 f (u) du =
£
¤
£
¤2
1
k
of agreement, given ρâ , is thus F (ρâ )2k + k1 1 − F (ρâ )k or, with
k 1 − F (ρâ ) . The overall probability
£
¤2
F = F (ρâ ), P (k, F ) = F 2k + k1 1 − F k .
Consider now what happens when a new action ǎ gets tried. If ρǎ < ρâ , then it just is as if one action got
£
¤2
1
removed from A \ B t . The probability of agreement is then P ((k − 1), F ) = F 2(k−1) + (k−1)
1 − F (k−1) .
If, however, ρǎ > ρâ , then ǎ becomes the new best known action. Denote F̌ = F (ρǎ ), then the probability
£
¤2
1
of agreement becomes P ((k − 1), F̌ ) = F̌ 2(k−1) + (k−1)
1 − F̌ (k−1) with F̌ ≥ F .
¤2
£
1
Combining these equations implies that we need to show that ∆P = F̌ 2(k−1) + (k−1)
1 − F̌ (k−1) −
³
£
¤2 ´
F 2k + k1 1 − F k
≥ 0 for k ≥ 2, F, F̌ ∈ [ 0, 1 ] and F̌ ≥ F . A long and tedious analysis shows that
this holds.8
¥
Note that this result assumes that the environment is static and that there is no turnover. Especially with change, we will have to be more careful on how to formulate the result although the basic
ideas will remain valid. The aspects change and turnover are briefly touched upon in section 4.6
8
The formal proof of this last claim is available from the author.
8
3.4
Communication and socialization
The basic model assumes that all members of the organization perfectly observe all realizations.
This is generally not the case. The flow of information about current and past actions and results
depends on the organization’s investments in communication and socialization. Such investments
may take different forms. Employees may, for example, be involved in decision making to make them
more aware of the decisions, trade-offs, and results. New members may be heavily socialized, with
stories about the organization’s great successes. A firm may be quite explicit about best practice
and ‘the way we do things around here.’ It follows that such investments may be an important
determinant of homogeneity. To study this, I relax here the assumption that all employees costlessly
observe the organization’s actions and outcomes. Instead, explicit investments are necessary to
communicate that information. Such investments, however, have no value unless these employees
are involved in decision making.9 So I will relax also the assumption that only the manager can take
actions, and will allow all employees to undertake actions that affect the organization’s performance.
In such world, which organizations will invest most in communication and socialization, and
thus become most homogenous? First of all, organizations in which employee decisions are important, since these are the ones that benefit most from employees having the ‘right’ beliefs. Second,
organizations that have experience that is of value to their employees. So I conjecture that organizations with high employee involvement that have been successful in the past will develop very
homogenous beliefs.
To study this formally, I focus here on a fairly specific extension of the basic model. Some of
the assumptions are driven by analytical considerations.
1. All members of the organization simultaneously undertake actions in each period. The organization’s payoff is (1 − β) times the payoff of the manager plus β times the average payoff
of the other members.
2. The manager maximizes the organization’s overall discounted performance. The other members just maximize their own performance.
3. Actions are unobservable (but each agent knows of course his own actions). None of the
members other than the manager directly observes any results. The manager observes her
own performance but not that of the other members.10
4. At the end of period 1, the manager can recommend her first period action to the other
members. Such communication costs the organization c > 0.
The results seem to hold without the more specific assumptions, but the analysis becomes very
complex. Note also that a ‘recommendation’ is just a convention, so that there can be two equilibria:
one in which employees interpret the manager’s ‘recommendation’ as meaning that the action is
‘good’ and another equilibrium in which employees interpret a ‘recommendation’ as meaning that
the action is particularly ‘bad.’11 I focus here on the equilibrium in which employees interpret a
recommendation as meaning that ‘the action is good’ since is the most logical, fits reality best, and
9
These investments also would have value if the employees had to expend private effort that is complementary, in
their payoff, to the probability of success of the course of action. I will not consider that case here.
10
Alternatively, we could assume that the manager also observed aggregate performance, but that seems to complicate the analysis enormously without affecting the qualitative results.
11
There is of course also a babbling equilibrium, which is equivalent to excluding the possibility of communication.
9
seems to Pareto-dominate. Lee and Van den Steen (2003) show why communicating ‘best practice’
is generally more valuable than communicating ‘worst failure.’
Let the return from the first period action be R.
Proposition 3 The probability of investment increases in β and in R.
Proof : I drop the superscript references to the period (all r’s are prior beliefs). Let, wlog., the manager’s
action in the first period be a1 .
Let V̂ denote the expected per-period payoff of the manager’s actions, from the manager’s perspective at
the start of period 2. Let V 0 denote the total expected per-period payoff, from the manager’s perspective at
the start of period 2, when the manager does not communicate any information. Without communication,
employee
actions are essentially random from the manager’s perspective. Therefore V 0 = (1 − β)V̂ +
P
R+
N
r
i=2 ai ,m
β
.
N
Let P denote the ex-ante probability that the employee will undertake a1 when the manager recommends it.
There are two equilibria. In one, P < N1 and the recommendation is interpreted as bad. In the other P > N1
and the recommendation is interpreted as good. As stated earlier, I focus on the second, so that P ≥ N1 .
Let V 1 denote the expected per-period payoff from the manager’sPperspective at the start of period 2,
N r ,m
when the manager does communicate. V 1 = (1 − β)V̂ + βRP + β i=2 Nai−1
(1 − P ) so that V 1 − V 0 =
³
¢
PN rai ,m ´ ¡
β R − i=2 N −1 P − N1 so that the manager will communicate iff
Ã
N
X
rai ,m
R−
N
−1
i=2
!µ
P−
1
N
¶
≥
c
β
(2)
The probability of communication thus increases in R.
PN r ,m
Let v, with law µ, denote the random variable v = i=2 Nai−1
. (Note that these are order statistics.) Let
α = β P c− 1 , then equation (2) will hold with equality for R = α + v. If the employee has a prior ra1 ,i about
( N)
R R ∞ ug(ra1 ,i ,u) du
dµ(v) which
a1 , then his expected value of R conditional on communication is R̃(ra1 ,i ) = v Rv+α
∞
g(ra ,i ,u) du
v+α
monotonically decreases in P on [ N1 , 1]. Moreover, as P ↓
(for given ra1 ,i ) as P → 1.
1
N,
1
R̃ ↑ ∞, while R̃ converges to some finite value
The employee will undertake a1 upon the manager’s recommendation iff R̃(ra1 ,i ) ≥ maxn≥2 ran ,i so P =
´N −1
R ³
f (u) du. The left hand side evidently strictly increases in P on [ N1 , 1], from N1 to 1. The
F R̃(u)
right hand side monotonely decreases in P on [ N1 , 1] from 1 to some value in [ 0, 1 ]. It follows that there
is a unique solution. This point defines the equilibrium and has P ∈ ( N1 , 1) and R̃ finite. Moreover, as β
increases, the right hand side decreases which implies that the probability of investment increases.
¥
The proposition thus shows that beliefs will be more homogenous in successful organizations and
when more of the organization’s members are involved in decision making
3.5
Employee involvement
There is an even more striking result with respect to employee involvement. If all members choose
actions and all actions and results are observed by everyone in the organization, then there will
eventually be perfect consensus, in the sense that all members of the organization choose identically
the same action and agree on its performance.
The formal assumptions of the model are the same as in the last section, except for the observability.
1. All agents simultaneously undertake actions in each period. The organization’s payoff is
(1 − β) times the payoff of the manager plus β times the average payoff of the other members.
10
2. The manager maximizes the organization’s overall discounted performance. The other members maximize their own performance.
3. All actions and results are perfectly observed by all members of the organization. All members
know each others’ beliefs.12
Let a sample point ω̂ denote a realization of all randomness, including mixed strategies.
Proposition 4 In any Markov-perfect equilibrium, there exists almost surely a T ∗ (ω̂), such that
after T ∗ all agents undertake the same action a∗ (ω̂) forever after. A Markov-perfect equilibrium
always exists.
Proof : I first show that a Markov-perfect equilibrium always exists. I do the proof by induction on the
size of the set of actions that have been tried, which I denote B. Let for any state S, b(S) = #B.
Note first that when b(S) = N (i.e. the set A \ B = B c is empty), it is optimal for all agents to play
â = argmaxa∈B ρa in every period that follows. So, for all states with b(S) = N , the strategies and value
function are well-defined.
Assume now that the value function and continuation strategies are well defined for all states in which
b(S) = k ≥ m. Consider a state S with b(S) = m − 1. Consider now the normal form game with the same
set of players, each with B c ∪ â as the action set, and the following payoffs. Whenever any agent chooses
an action in B c (so that the state will transition to a state with b(S) = k with k ≥ m) each agent gets the
respective immediate payoff of his action plus the continuation payoff which follows from the induction step.
ρâ
When all agents choose â, the payoffs to all are (1−δ)
. Pick any (possibly mixed-strategy) equilibrium of
this normal form game (of which there exists at least one) and define these as the equilibrium strategy for
this state. These strategies are well-defined and clearly Markov, and it is straightforward to check that this
is indeed an equilibrium. The value functions follow.
For the first part of the proposition, consider a period in which the state is S with b(S) = k. If, in that
period, the agents undertake m (distinct) unknown actions a ∈ B c , then the state transitions to some S 0
with b(S) = k + m. Consider now some state S. If the strategy of at least one player, say i, is to play
some unknown action ãi ∈ B c with at least some probability, then the state will almost surely transition at
some point to a state S 0 with b(S 0 ) > b(S). Since b(S) ≤ N ∀N there can be only a finite number of such
transitions, so that, almost surely, after some time only known actions get played. If the strategy is for all
players to play â for sure, then the state remains the same for the next period, so that the same strategy is
optimal forever after. It follows that all players play â forever after. This then completes the proposition.
¥
This is actually a very strong result. In the end, all members of the same organization will come
to act identically. When only the manager takes actions it often happens that an action never
gets tried even though some agent is convinced that that action dominates all others. In that
case, disagreement on the optimal action persists. When each agent is free to pursue his or her
preferred actions, such disagreement cannot persist. This does still not imply that the members of
the organization will try all actions. So it will still happen that their eventual action is not optimal
and that different organizations end up with different eventual actions.
Overall, this section has showed that organizations have a natural tendency to develop homogeneity and identified some key factors that play a role in this process. The next section links this
natural homogeneity to the concept of corporate culture.
12
The assumption that they know each others’ beliefs simplifies the analysis considerably, but is not necessary.
11
4
Homogeneity and Corporate Culture
The importance of shared beliefs has not escaped managers or management theorists. Thomas
Watson Jr., the legendary manager of IBM, already famously stated that ‘I firmly believe that any
organization, in order to survive and achieve success, must have a sound set of beliefs on which it
premises all its policies and actions’ (Watson 1963). More important, seminal authors on corporate
culture, such as Burns and Stalker (1961), Schwartz and Davis (1981), Peters and Waterman (1982),
Donaldson and Lorsch (1983), and especially Schein (1985) and Kotter and Heskett (1992), have
defined shared beliefs as an essential part of corporate culture. While other ways of defining culture
may be useful, as discussed in appendix A, this clearly suggests shared beliefs are an important
part of culture.
The authors who defined culture as shared beliefs have also suggested a pattern of ‘facts’,
based on their case studies (Donaldson and Lorsch 1983, Schein 1985, Kotter and Heskett 1992).
These ‘facts’ are more or less as follows. Organizations in similar circumstances may develop very
different cultures, and these cultures may persist even in the view of complete turnover and even
when there are clear signs that other cultures may be more effective. An organization’s culture
is often determined by the beliefs of the original leader and by early experiences, and external
succession of the CEO is more likely to lead to a change in culture than internal succession. Strong
cultures will be found among older organizations.
The purpose of this section is twofold. First of all, I want to verify that this pattern of ‘facts’
does indeed hold in the current model. This serves both as a formalization and potential formal
explanation of these informally observed ‘facts’ and as a partial validation of the model. Second, I
derive some new implications for corporate culture, in particular the fact that slow feedback from
the environment will increase the heterogeneity of cultures across firms and the influence of the
CEO’s original beliefs.
Since the tractability of the model is much higher with a single decision maker, I will focus on
that model and identify the organization’s eventual action with its culture. The reason is that in the
case of a single decision maker, the eventual action is also the one on which most employees agree as
the optimal action. The results seem to extend, in an appropriate sense, to more general models. I
will also use the earlier measures of homogeneity as measures for the ‘strength’ of corporate culture,
as discussed in appendix A.
4.1
Diversity and origin of culture
The first ‘fact’ is that otherwise similar organizations may develop very different cultures. I obtain
an even stronger result in this model: as the number of alternative actions increases, the probability
that two organizations have different cultures converges to 1. To see this formally, let a∗f denote
organization f ’s eventual action or culture.
Proposition 5 As N → ∞, P [a∗f = a∗g ] → 0.
PN
Note that P [a∗f = a∗g ] = n=1 P [a∗f = an ]P [a∗g = an | a∗f = an ]
PN
PN
= N1 n=1 P [a∗g = an | a∗f = an ] ≤ N1 n=1 P [a∗g = argmaxa∈A ρa ] = P [a∗g = argmaxa∈A ρa ]. By proposition 6, P [a∗g = argmaxa∈A ρa ] → 0 as N → ∞, so that the proposition follows.
¥
Proof :
12
Dysfunctional cultures The above discussion suggests that the manager does not necessarily
settle on the optimal action. In fact, when the number of actions increases, the probability that the
manager settles on the optimal action goes to zero. To see this formally, let âN = argmaxa∈A ρa
when A has N elements.
Proposition 6 As N → ∞, P [a∗ = âN ] → 0.
N
Proof : Fix a given set of potential returns {ρan }∞
n=1 with only the set AN = {an }n=1 available to the
firm. Let âN = argmaxa∈AN ρa , which is unique for all N with probability one (which I assume henceforth).
Since P [âN = a∗ ] ≤ P [âN ∈ B] (where B is the set of eventually tried actions), it suffices to show
that P [âN ∈ B] → 0 as N → ∞. The probability that âN gets tried equals the probability that
there is no action ã that (1) has a prior rã,m ≥ râN ,m and (2) has a true value ρã such that râN ,m ≤
R∞
δ
ρã − (1−δ)
(u − ρã )g(râN ,m , u) du. We can write the latter condition as ρã ≥ h(râN ,m ) for some increasing
ρã
function h. Given some râN ,m , the probability that ã satisfies conditions 1 and 2 is P [rã,m ≥ râN ,m & ρã ≥
h(râN ,m )] = (1 − F (râN ,m ))Iρã ≥h(râN ,m ) so that, still for given râN ,m , the probability that no action satisfies
h
i
them is Πa∈AN \âN 1 − (1 − F (râN ,m ))Iρa ≥h(râN ,m ) . Since râN ,m is drawn from F , we get P [âN ∈ B] =
£
¤
R
Πa∈AN \âN 1 − (1 − F (u))Iρa ≥h(u) f (u)hdu. When we go from N ito Nh+ 1 actions, the integrand gets
muli
tiplied with an extra factor, being either 1 − (1 − F (u))IρâN ≥h(u) or 1 − (1 − F (u))IρaN +1 ≥h(u) . Either
way, with probability one (over the realizations of returns {ρan }∞
n=1 ), the integrand converges to zero for any
u, so that P [a∗ = âN ] → 0. That proves the proposition.
¥
It is thus clear that organizations can have dysfunctional cultures in this model. Moreover, the
organization’s manager will be aware of the fact that his culture is almost surely suboptimal.
There is another important sense in which cultures in this model can be dysfunctional. While
I implicitly treated the payoffs as if they were also the firm’s profits, there is nothing in the model
that requires this to be the case. If the agents’ payoffs differ from these of the firm, we can get very
dysfunctional cultures that may eventually destroy the organization. Enron managers, for example,
learned that the easy way to success was deceiving shareholders and regulators, and developed a
culture that allowed such behavior, which eventually destroyed the firm. NASA engineers learned
that things went smoother if they minimized potential problems and just hoped for the best. A
culture developed that allowed such behavior to flourish, with disastrous consequences.
The fact that similar organizations can develop very different cultures and that cultures can
even be dysfunctional raises the question what factors determine a firm’s culture.
4.2
Managerial beliefs as a determinant of culture
The management literature suggests that an organization’s culture is influenced by the beliefs of
founders and early managers (Donaldson and Lorsch 1983, Schein 1985, Kotter and Heskett 1992,
Baron et al. 1999). The following proposition confirms this effect in the current model. It shows
essentially that a certain way of doing things is more likely to become an organization’s eventual
course of action if the manager originally is very positive about it.13
Proposition 7 For any action ǎ, the probability that a∗ = ǎ increases in rǎ,m .
13
An alternative perspective of a manager’s influence on culture is presented in Van den Steen (2002d). In that
model, a manager’s strong beliefs or vision cause sorting in the labor market, which leads to a homogeneity in the
firm’s beliefs.
13
Proof : The probability that ǎ gets tried increases in rǎ,m since it equals the probability that there is no
R∞
δ
action ã that (1) has a prior rã,m ≥ rǎ,m and (2) has a true value ρã such that rǎ,m ≤ ρã − (1−δ)
(u −
ρã
ρã )g(rǎ,m , u) du. Conditional on getting tried at least once, all actions are ex ante equally likely to become
the eventual action a∗ . This implies the proposition..
¥
It is also trivial but intriguing that (ex-post) the most influential managers are those under whose
early actions the firm was very successful. This corresponds well with the observation that in many,
if not most, case studies of firms with a strong culture, the firm had an early leader who had strong
beliefs and was successful.
4.3
Managerial succession and persistence of culture
Given now the manager’s important influence on a firm’s culture, what happens when the firm
changes managers? The model suggests two results.
1. The culture does not disappear with the manager. In particular, if an insider succeeds the
manager, then the culture is likely to be preserved.
2. When an outsider succeeds the manager, however, the culture is more likely to change than
with an insider.
These conclusions resonate well with the case studies of management theorists such as Donaldson
and Lorsch (1983) or Kotter and Heskett (1992) who have stressed the persistence of culture and
the need to bring in new managers as the key step in changing a firm’s culture.
To study this formally, I need to extend the model to allow for succession. Assume that at the
end of each period, there is a probability p that the manager will be replaced by a new one. Assume
that a manager only cares about the organization’s performance under her own management, and
that therefore the probability p is just a factor in the manager’s discount factor δ. I will compare
the case that the successor is one of the firm’s current members, i.e. an insider-successor to the
case in which the successor is a member of another organization, i.e. an outsider-successor. Insiders
know the organization’s history in terms of actions and outcomes. Outsiders have no information
about this organization’s prior actions and outcomes but do have an equal amount of experience
as a member of another randomly drawn firm.
Since I want to compare eventual actions, I will assume that the period t ≥ N . Let a∗1 , a∗insider ,
∗
aoutsider denote the eventual action under respectively the original manager, the insider-successor
and the outsider-successor.
Proposition 8 The probability P [a∗insider = a∗1 ] ≥ P [a∗outsider = a∗1 ] Moreover, in the limit as
N → ∞, P [a∗insider = a∗1 ] → x ≥ 1/2 while P [a∗outsider = a∗1 ] → 0.
Proof : Denote by a∗2 , the eventual action of the manager of the outsider-successor’s original organization.
For the first part of the proposition, fix a set of identical prior beliefs for the insider and outsider successor
and a set of ρa ’s. Note that, if they started from scratch, they would arrive at the same eventual action,
say a∗ . Consider now first the case that a∗ = a∗1 . While the outsider may settle on either a∗1 or a∗2 , the
insider will always settle on a∗1 in the original game. Consider next the case that a∗ 6= a∗1 . The insider will
eventually settle on either a∗1 or on a∗ . If a∗2 6= a∗1 , then the outsider will never settle on a∗1 . If a∗2 = a∗1 ,
then the likelihood that the outsider settles on a∗1 is identical to the likelihood that the insider settles on a∗1 .
Overall, it follows that the insider is always at least as likely to settle on a∗1 as the outsider.
For the limit results, consider first the case of the insider successor, denoted I. Note that ∀a ∈ A, P [ρa∗1 ≥
ρa ] ≥ 1/2. To see why, note that if a ∈ B, then ρa∗1 ≥ ρa for sure. If a 6∈ B, then ρa is distributed according
14
to F , while ρa∗1 is distributed according to F #B , which first order stochastically dominates F . Consider then
the modified problem that I’s choice set is restricted to A \ a∗1 . Let a∗ denote I’s eventual action in this
modified problem. In the unmodified problem I’s eventual action a∗insider must then be either a∗1 or a∗ .
Note that I’s eventual action can be a∗ only if ρa∗1 ≤ ρa∗ . Furthermore, P [ρa∗1 ≥ ρa∗ ] ≥ 0.5 by the earlier
argument. So it follows that the probability that a∗insider = a∗1 must be weakly larger than 1/2 in the limit.
Consider next the outsider-successor. Since there is no information transfer, it is as if we consider a similar
succession in a completely different organization. An immediate extension of proposition 5 implies then that
as N → ∞, the probability of the eventual actions being identical goes to zero.
¥
A change in culture is thus more likely under an outsider than under an insider successor. Simulations suggest that this effect can be large. Table 1 gives the results of 5 simulations with 50
organizations each, in which one manager runs the organization for N periods and then gets succeeded either by one of her organization’s own members or by an outsider. The numbers in the
table represent the percentage of cases in which the successor eventually chose the same action as
the original manager. Clearly, outsiders are much more likely to change the culture than insiders.
The effect may have significant implications, as can be seen from the average performance difference
column.
Simulation
Insider
Outsider
1
2
3
4
5
68
66
72
64
78
10
12
14
18
10
Avg. % Performance difference
24
19
20
14
20
Table 1: Percentage of cases in which the eventual action of the successor-CEO is identical to the
eventual action of the original CEO, and the average percentage performance difference. The data
represent 5 simulations of 50 organizations each. The number of actions was 100.
The effect is essentially one of forced learning: during her tenure, the original manager implicitly
chooses what her successor learns. When the successor takes over, he might try out a few changes,
but if he doesn’t quickly find an action that performs really well, he will fall back on the proven
strategy of his predecessor. Combined with the earlier result, this implies that a manager’s beliefs
may determine an organization’s culture even after the manager is gone.
This also suggests a prediction on CEO succession: a rationally acting board of an underperforming firm should select an outsider as successor to a retiring CEO, while the board of an
excellent performer should select an insider.
More on the persistence of culture One of the striking things about corporate culture is its
persistence over time. Even after all the original members are gone, the organization may still
have the same shared beliefs. It is as if the organization has its own personality, independent of
individual members.
The above analysis suggests one reason for such persistence.14 The simulation in table 2 puts
more flesh on this idea. It shows the results of simulations of multiple consecutive successions, such
that the original manager-founder and the last manager never overlapped. Consider in particular
14
A different mechanism is of course sorting, as discussed in Van den Steen (2002a).
15
the following simulation. A manager runs the organization for N periods. At that point one of
her organization’s members succeeds her, while all other members get replaced by new ones who
did not observe the organization’s earlier history and who do not make inferences about it.15 After
another N periods, the second manager is succeeded by one of the organization’s new members.
Note that the original manager had left the organization by the time this last manager entered.
The results in the table indicate for 5 simulations how often the last manager settles on the same
action as the first manager. Although they never met, the first manager’s beliefs influence the final
manager’s actions.
To control for the fact that this might be due to convergence on an optimal action, I ran the
simulation simultaneously for two organizations and report in the second and the third column the
percentage of cases that the two organizations end up with the same action and their performance
difference. Clearly something different than convergence on an optimal action is going on.
Simulation
1
2
3
4
5
Culture
tence
Persis60
58
56
56
72
Identical to other
organization
20
22
16
16
18
Avg. % Performance Difference
18
15
15
17
18
Table 2: Percentage of cases in which the eventual action under the final manager is identical to the
eventual action under the original manager, and percentage that the eventual action is identical to
that of some other organization. The data represent 5 simulations of 50 organizations each. The
number of actions was 100.
4.4
Rate of learning
Apart from formalizing and formally explaining the informally observed pattern of ‘facts’, we can
also derive new insights from the model. An interesting one is the effect of the rate of learning. A
natural conjecture would be that in more fuzzy environments, cultures would be more diverse and
the influence of the manager’s beliefs would be stronger.
The current model allows a comparative static that partially captures this. Consider in particular the following interpretation of the model. Assume that the manager can change actions at
any point in (continuous) time, but she has to wait at least the duration of a period, denoted ∆,
to learn an action’s payoff. In any optimal strategy the manager will only change actions at the
discrete time points t∆. Since the rate of learning is determined by the length of ∆, and the length
of ∆ is reflected in the discrete-time discount rate δ (for a given continuous-time discount rate) , δ
captures that rate of learning.
The intuition is now that in an environment where it takes longer to learn about an action’s
performance, cultures would be more diverse and the influence of the manager’s original beliefs will
be stronger. To see this formally, let a∗f and a∗g denote the eventual cultures of randomly selected
organizations f and g respectively.
Proposition 9 The probability P [a∗f = a∗g ] increases in the discount rate δ. The probability that an
organization’s eventual culture is the one originally preferred by the manager, P [a∗ = argmaxa ra ],
15
I make this assumption for analytical reasons. Taking into account such inferences would strengthen the result.
16
decreases in δ.
Proof :
For given (ρa )a∈A and {ra,m }a∈A , equation (1) says that (in a set-wise ordering sense) the set of
actions Bit that have been tried by agent i’s manager by the start of period t, increases in δ.
Fix (ρa )a∈A . Assume, essentially wlog., that all the ρa are different.16 Let then, wlog., the actions be
numbered such that ρan ≥ ρan+1 . Let bk,i denote the event that ‘a∗i = ak ’, which is also the event that
‘ak ∈ Bi & ∀l < k al 6∈ Bi ’, where Bi is the set of actions that are eventually tried by the manager of agent
i. Note that, conditional on (ρa )a∈A and with 1 and 2 in different organizations f and g, bn,1 and bn,2 are
PN
PN
2
independent. We have thus that P [a∗1 = a∗2 ] =
n=1 P [bn,1 ∩ bn,2 ] =
n=1 P [bn,1 ] . Since I can focus
completely on one organization now, I will drop the reference to the agent. Let bn and b̌n denote a∗ = an
when the discount factor is respectively δ and δ̌ = δ + ² with ² ≥ 0. I will argue that (1) P [bn ] ≥ P [bn+1 ]
and idem for b̌n , and (2) ∪kn=1 bn ⊂ ∪kn=1 b̌n . For point (1), note that when ρa(n+1) = ρan then, by symmetry,
P [bn+1 ] = P [bn ]. Point (1) then follows from the fact that P [bn ] increases when ρan increases. For point (2),
note that B ⊂ B̌ and that ∪kn=1 bn are exactly all the sample points for which one or more actions an with
n ≤ k will eventually have been tried. So ∪kn=1 bn = {ω : ∃n ≤ k with an ∈ B} = ∪n≤k {ω : an ∈ B} For any
ω for which an ∈ B, evidently an ∈ B̌, so that ∪n≤k {ω : an ∈ B} ⊂ ∪n≤k {ω : an ∈ B̌} or ∪ki=n bn ⊂ ∪kn=1 b̌n .
This proves point (2).
By point (2), it follows that there exists a set of events ‘αn,m = bn ∩ b̌m ’. For m < n, event αn,m denotes
the sample points that moved from n to m when the discount P
factor went fromP
δ to δ̌. We can write
b̌m = bm ∪ (∪n>m αn,m ) \ (∪n<m αm,n ) so that P [b̌m ] = P [bm ] + n>m P [αn,m ] − n<m P [αm,n ]. It then
follows that
(
"
"
#)
#2
N
N
X
X
X
X
X
X
2
∗
∗
P [bm ] + 2P [bm ]
P [a1 = a2 | δ̌] =
P [αn,m ] −
P [αm,n ]
+
P [αn,m ] −
P [αm,n ]
n>m
m=1
n<m
m=1
n>m
n<m
Since P [αn,m ]/P [bm ] and P [αm,n ]/P [bm ] tend to zero as ² → 0, the last term will be dominated in the limit
as δ̂ → δ by both other terms. Then we can write
P [a∗1
=
a∗2
| δ̌] =
P [a∗1
=
a∗2
| δ] +
N
X
"
2P [bm ]
n>m
m=1
= P [a∗1 = a∗2 | δ] +
X
N
N
X
X
P [αn,m ] −
X
#
P [αm,n ] + o(. . .)
n<m
2 [P [bm ] − P [bn ]] P [αn,m ] + o(. . .)
m=1 n=m+1
Note now that P [bm ] ≥ P [bn ] in that equation since n > m (and the values go up with lower indices). It
follows that P [a∗1 = a∗2 | δ̌] ≥ P [a∗1 = a∗2 | δ]. This completes the first part of the proposition.
For the second part, let again the (ρa )a∈A be given and fix the manager’s prior. Let ã be the action that
0
the manager originally thinks is best, i.e. ã = argmaxa∈A ra,m
. Let again δ̌ > δ, so that for any ω, B ⊂ B̌.
Since the manager originally thinks ã is best, ã ∈ B and ã ∈ B̌. If ∀a ∈ B̌ ρã ≥ ρa , then ∀a ∈ B ρã ≥ ρa .
This proves the proposition.
¥
4.5
Culture and performance
Historically, the interest in corporate culture has been largely driven by its suggested impact on corporate performance. In particular, works such as Deal and Kennedy (1982), Peters and Waterman
16
With identical ρa , the analysis still goes through, but it becomes messy since there is no transparent notation.
17
(1982), or Collins and Porras (1994) have popularized the notion that culture is a driver of performance. Most economic analyses of culture have essentially tried to explain the benefits of culture
(Kreps 1990, Crémer 1993), although they do explicitly admit to potential negative side-effects.
It is unclear, however, whether this popular notion is really correct. The case studies in these
management books (and our casual observations) are subject to important selection biases. More
systematic studies are rare and the three systematic studies that are often mentioned in the literature (Kotter and Heskett 1992, Burt, Gabbay, Holt, and Moran 1994, Sørensen 2002) use the same
rudimentary data set.17
The position of this paper is that ‘culture happens’ and whether that is good or bad for the
organization remains to be seen. Since homogeneity reduces the incentives to collect information,
which may hinder innovation and adaptation, culture may sometimes be plainly bad.
Nevertheless, the theory of this paper does predict a correlation between culture and performance, but it is performance that causes homogeneity, and not the other way around. The
simulation result in figure 1 suggests that this correlation may be quite strong. It depicts the result
of a simulation where a manager takes actions until she settles on some a∗ . At that point the
manager decides whether to communicate a∗ and its payoff to the members of the organization,
who take this information then as given, without any further inferences.
Relation between strenth of culture and performance
4
3.5
3
Performance
2.5
2
1.5
1
0.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Strength of Corporate Culture (Probability of agreement on optimal action)
1
Figure 1: Performance in function of homogeneity (measured as probability of agreement on optimal
action). The simulation consisted of 100 organizations with 500 members each, choosing among 10
arms.
With such correlation, it would be understandable that many people get struck by the strong
cultures of extreme performers. They are led to conclude that culture must be one of the keys to
their success and then try to develop theories to explain this obvious ‘fact’.
This result thus questions the inferences that can be drawn from simple regression analyses,
such as these by Kotter and Heskett (1992), and the ‘received wisdom’ on the relationship between
culture and performance.
17
Lazear (1995) argues that the most famous, Kotter and Heskett (1992), also has important issues with the
analysis.
18
4.6
Other aspects of culture
There are many important aspects of culture that I have not dealt with in this paper, but for which
the model could have interesting implications.
Culture and turnover Turnover should weaken the culture, since new members do not possess
the same experience as existing members. The situation is more complex than that, however.
For one thing, new members will learn from the organization’s experience by deducing information
from the organization’s actions, by (imperfectly) observing the beliefs of colleagues, and from stories
about the past successes and failures of the organization. There will also be sorting in the turnover
and in the hiring process, which would typically increase homogeneity (Schein 1985, Van den
Steen 2002a). Finally, the impact of turnover on communication and socialization investments is
ambiguous: while the effect of socialization is likely higher with new members, the payoff from
these investments are lower when members stay shorter.
The role of turnover is thus clearly a research topic on itself. Note that some firms with
extremely high turnover are notorious for their strong cultures. Consulting is a case in point.
Culture and Change Culture is often mentioned as an impediment to change. In the current
model there does not seem to be such effect, except for a potential spurious relationship in which the
belief strength of the organization’s members causes both homogeneity and resistance to change.
Introducing change in the model could, however, affect the results quite profoundly.
Variability of performance Sørensen (2002) argues that a strong culture should reduce the
variability of performance and shows that this holds indeed in the data set of Kotter and Heskett
(1992). The current model suggests at least one causal link for such relationship: if members are
randomly selected to undertake actions, more homogenous beliefs about what to do will definitely
cause more systematic behavior.
Distinctiveness of culture The analysis in this paper has focused on internal homogeneity. To
understand the importance of homogeneity and culture, however, it would be useful to compare
this to the heterogeneity of beliefs across organizations.
5
Testing strategies
The idea that experience and learning may generate homogenous beliefs has been suggested in the
context of case studies, of which Donaldson and Lorsch (1983) and Schein (1985) are probably the
most famous. Schein (1985) includes this aspect even explicitly in his most formal definition of
culture, as discussed in appendix A. While this is encouraging, it is certainly a long way from
systematic empirical evidence. Moreover, while these stories fit quite well the idea that firms with
a successful history and high employee involvement have the most homogenous beliefs, there is no
systematic research on these comparative statics. There is thus a great need for more empirical
analysis of homogeneity.
The most direct test of the theory is to measure people’s beliefs within and across organizations
and test for the relations predicted by the model. The key issue is obviously to find a good
instrument for measuring beliefs. While not very common in economics, methods for measuring
beliefs have been developed and validated in psychology.
19
An alternative approach is to take actions as proxies for beliefs and thus test whether behavior
is more similar within than across organizations, and whether such similarities relate indeed to
things like past performance and age of the organization.
There are also some indirect tests, although the potential for alternative explanations is larger.
One possibility is to test whether firms change more under outsider-successors than under insidersuccessors, or that badly performing firms are more likely to hire outsider successors to the CEO.
Another possibility is to test whether people who worked for the same organization tend to act in
similar ways after they leave. The latter could be tested in the style of Bertrand and Schoar (2001).
The traditional difficulty of separating learning from sorting may be less of an issue in this context since we are ultimately interested in homogeneity and these are simply substitute mechanisms
(Van den Steen 2002a).
A final possibility are experiments in the style of Weber and Camerer (2001). A possible setup
would have groups of people repeatedly play a (partially random) game and see whether groups
develop internally homogenous beliefs about the optimal actions and whether such beliefs can
persist over generations of players.
6
Conclusion
Since homogeneity of beliefs may be an important determinant of agency problems, understanding
its origins is important to understand the trade-off between markets and organizations and to decide
on incentive systems and organization design.
This paper pointed out this very important role of homogeneity and showed that members of the
same organization naturally develop such homogeneity through shared experiences. It showed that
such homogeneity will be especially pronounced in older organizations with a successful history and
high employee involvement. A number of factors play a role in this phenomenon: the organization
is permanent, so that there is a considerable number of events over which the members can share
experiences; the information on actions and results gets shared within the organization since it
is to the advantage of the organization to have its members understand the sources of success
and failure; the information spreads less easily to other organizations since the focal organization
usually has incentives to keep the information private and since it is often difficult to communicate
the information, except through direct observation or repeated rich stories.
I then connected this homogeneity to the notion of corporate culture and show how the model
formalizes a pattern of ‘facts’ that have been informally suggested by the literature on culture. The
analysis thus suggests an explanation for these ‘facts’ while at the same time partially validating
the model. I also showed that a lower rate of learning leads to more diversity of cultures and
increases the impact of the founder. I finally questioned the received interpretation of the correlation
between culture and performance, based on the earlier conclusion that strong performance will cause
homogeneity.
Although there is some informal evidence that lends support to the theories of this paper, there
a clear lack of systematic empirical evidence. This should be a priority area for future research.
This paper seems to have only scratched the surface. Lots more questions remain on both the
causes and the effects of homogeneity.
20
A
Corporate Culture
When coming into contact with an organization, people are often struck by the fact that members of the
organization seem to act and think similarly, but differently from members of similar other organizations.
It is as if each organization has its own ‘personality.’ Moreover, this ‘personality’ may remain remarkably
constant over time. Even when many of the original members are gone, the new generation thinks and acts
in very much the same way as their predecessors. It is essentially this character of an organization, which
some have more than others, that has been called its ‘culture.’ Given the rather vague phenomenon, it is not
surprising that there are many divergent definitions in the literature. Moreover, as the term became more
popular, it also began to live a life of its own. Lazear (1995) provides a survey of alternative definitions.
Rather than trying to replicate such survey here, the purpose of this section is to present the view in the
management literature on which this paper is based, and compare it to some key alternatives.
A.1
Examples of corporate culture
To fix ideas it is useful to start out with presenting examples, drawn from personal experience and case
descriptions.
The first example is a comparison between the Brussels offices of Arthur D. Little and McKinsey in the
mid to late nineties. These local offices served similar clients, were started at about the same time and were
similar both in size and in personnel composition.
Arthur D. Little’s consultants proudly stated that their firm was an organized chaos or chaotic organization and that it had as many strategies as there are consultants. While formal training existed, every team
really went its own way. Data analysis was not so important, but listening to people was key. Conclusions
were often backed up by quotes from clients, or by stories. It was important to have an open mind and not to
come too quickly to conclusions. There were very few formatting standards for presentations. Performance
evaluations were done every few months via informal 5-minute chats. Every consultant was responsible for
his or her own staffing via a market-based system. People took lunch while working in their office. Team
lunches were exceptional. Arthur D. Little called itself ‘the Company.’
McKinsey’s well-developed consulting methodology, on the contrary, guided each study pretty closely.
A new study started by collecting all ‘knowledge’ about similar studies that had been conducted in the
past in other offices. From the start of the study, consultants were supposed to think in terms of final
client recommendations. Any conclusion had to be backed up by data. There were strict formats for the
presentations, decided upon by a global committee of senior directors. There was a clear one-firm policy:
the process, rules and systems should be similar all around the globe. Consultants got evaluations every 6 or
12 weeks, using extensive and formal evaluation forms. Staffing was centralized and future assignments were
chosen to improve on weaknesses. Consultants spent nearly all their time at the client site. Lunch (and for
out-of-office teams also dinner), were taken as much as possible with the team. McKinsey called itself ‘the
Firm.’
These were two firms that were essentially in the same business but worked in very different ways. There
were no obvious structural limitations or legacy systems that prevented one to switch to the other’s model.
Both were aware of the differences. In fact, these differences in behavior reflect differences in opinions and
beliefs among the most senior people about the relative importance of ‘the one best way of doing things,’
individual creativity, teamwork, the most effective process of doing consulting, etc. A telling fact is that new
consultants at McKinsey receive a copy of the book Perspective on McKinsey, by Marvin Bower, McKinsey’s
de facto founder. The book is accompanied by a memo from Bower, urging ‘not to give or loan copies to
people outside the Firm.’ The book essentially gives Bower’s perspective on the ‘lessons that I believe might
be learned from our successes, mistakes, and failures, [. . . ].’ Some McKinsey people refer to it as ‘the bible.’
It seems a conscious effort to influence the beliefs of new employees. The fact that it is explicitly for internal
use only, is a clear statement that this is not posturing for the outside world, but valuable information from
which new employees can and should learn.
Cultures also comes in less functional forms. Some companies, such as the former Enron, encourage
their people to be aggressive and push limits, even if it gets them close to legal limits. Other organizations,
including some government administrations, have implicit shared beliefs that initiative creates personal risks
21
without rewards. Some firms have a strong ‘nine to five’ culture while in others people always stay late,
even if they don’t have anything to do. Cultural differences can also relate to the importance of consensus,
the treatment of new employees, the level of confrontation, the level of cooperation, the competitiveness, the
implicit importance and status of engineers versus marketers, the existence of reserved parking spots for top
management, open or closed door policies, etc.
Note that cultures can also develop along other dimensions than firms. We can talk, for example, about
a sales culture versus a production culture, or about the culture of academic economists as opposed to that of
academic sociologists or engineers. Each of these groups have a set of common experiences they go through,
and develop a set of shared beliefs.
A.2
Definition of corporate culture and the role of learning
Since culture is a complex social phenomenon, it has multiple dimensions and therefore multiple potential
definitions, that all have their value in the right context. While I mention some alternative definitions below,
this paper uses what seems to be the most prevalent definition in the management literature: corporate
culture as shared values, beliefs, and assumptions which generate behavioral norms and ‘the way we do
things around here.’ Note that the management literature often uses the terms values and beliefs almost
interchangeable.18
This idea of culture as shared beliefs or values goes back at least to Burns and Stalker (1961) who, in their
seminal discussion of ‘organic’ versus ‘mechanistic’ organizations, define culture as ‘a dependable constant
system of shared beliefs.’ Other early contributions were the work of Baker (1980) and Schwartz and Davis
(1981) who defined culture respectively as ‘some interrelated set of beliefs, shared by most of their members’
and ‘a pattern of beliefs and expectations that is shared by the organization’s members.’ A key impetus in
popularizing the notion of culture was the bestseller In Search of Excellence by Peters and Waterman (1982),
who defined culture as ‘shared values’ but stress that they also mean ‘basic beliefs.’ Donaldson and Lorsch
(1983), which is often considered a seminal work on corporate culture, do not mention the word culture, but
talk instead about managerial beliefs. Most of these authors suggest that a culture can have subcultures.
Probably the most cited perspective on corporate culture is that of Schein (1985). He defines culture as
having three levels. The most visible, but most superficial, level is that of culture as a pattern of behavior.
It is ‘the way things are done around here,’ the norms, the stories, the symbols. These behavioral patterns
reflect a second, deeper, level of culture, which are the firm’s shared values. Shared values are on their
turn driven by the third and most fundamental level of culture: shared assumptions. Kotter and Heskett
(1992) base their definition on Schein (1985), but eliminate the distinction between beliefs and values. Note
that, although these authors focus on culture as shared beliefs, their ultimate interest is in the behavioral
implications of such theory. But to develop a systematic theory of that behavior, the idea is that we must
look deeper, to values and beliefs.
This literature that defines culture as shared beliefs and assumptions also often mentions the role of
learning. Schein (1985) deems the process of shared experience so crucial that he even includes it in his
definition of corporate culture as ‘a pattern of shared basic assumptions that the group learned as it solved
problems of external adaptation and internal integration, that has worked well enough to be considered valid
and, therefore, to be taught to new members as the correct way to perceive, think, and feel in relation to
those problems.’ Donaldson and Lorsch (1983) claim that ‘although the founders’ personal beliefs lie at the
heart of the belief system, corporate history also plays an important part in shaping current beliefs. As the
founders and their successors manage by their principles, their experiences lead them to modify the system
through the process of incremental change.’ Schwartz and Davis (1981) state that ‘culture reflects what has
worked in the past.’ In their analysis of how cultural change can occur, Kotter and Heskett (1992) observe
that ‘the importance of results cannot be overstated. These new cultures grew in a cycle that was driven
by successful results.’ All these remarks suggest that culture is developed to a large extent through joint
learning from the company’s experiences.
18
The issue is not unfamiliar to economists: you need some fairly strong assumptions to separate beliefs and utilities
into expected utility. More informally basis, valuing the environment is closely related to beliefs about where our
planet is headed.
22
As mentioned earlier, however, there are important alternative to this idea of culture as shared beliefs.19
An important alternative is the idea that culture is simply a set of conventions, arbitrary ‘ways we do things
around here’. This is similar to the first notion of culture in the Kreps-Hermalin model. As mentioned
earlier, this requires a shared belief about what exactly the coonvention is, and shared beliefs about the
world will facilitate the development of conventions.20 To other important alternatives are the notion of
culture as shared language or meaning and the notion of culture as group norms. (Note that a convention is
an arbitrary rule that solves a coordination problem while a norm has a moral implication.) Finally, Martin
(1992) suggests a ’fragmentation’ perspective that has a ‘focus on ambiguity as the essence of organizational
culture.’ Martin recognizes that ‘the Fragmentation perspective is difficult to discuss with clarity’ and
considers it to be founded in the postmodernist tradition. Some would call it an anti-theory, a theory about
the non-existence of theory. In her discussion, Martin explicitly states that the Integration perspective,
of which this paper is essentially part, ‘has become the dominant view of organizational researchers and
practitioners.’
A.3
Measuring Culture
When we define corporate culture as shared beliefs, there are essentially two interpretations for the notion
of the ‘strength’ of a culture:
1. Internal homogeneity: the degree to which people within the firm share the same beliefs.
2. External heterogeneity: the degree to which these common beliefs are different from the beliefs of the
population at large, or from the beliefs of other organizations.
To prevent confusion, some of the literature, reserves the term ‘strength’ for the first concept (i.e., for the
degree of homogeneity of beliefs within the firm). I will therefore use ‘distinctiveness’ to denote the second
concept, the degree of difference with the population at large.
A.4
The Difficulty of Communicating Information
The development of shared beliefs that differ from those of other organizations, as envisioned by this paper,
presumes that information and experience are more easily shared within than across organizations. It should
be uncontroversial that some information is indeed difficult to communicate. Try the following ‘thought
experiment.’ Observe a person on the street during one second. Now try to describe that person to someone
else in such detail that the other would be able to recognize the person with the same ease as you yourself
would. This would take a tremendous amount of time, if it is at all possible. This idea that some information
is difficult to communicate is well captured by Confucius’ ‘Tell me and I will forget, Show me and I will
remember, Involve me and I will understand!’ This is further compounded by the incentives for organizations
to keep their information and experience private, especially in competitive circumstances. Note also that
transfer of information is very difficult to contract on.
Within the firm the situation is very different. The competitive concerns are much less. More importantly,
employees are active participators in and direct observers of the firm’s action and outcomes. Communication
with existing employees happens thus to some extent automatically, although the firm can still manage this
by, for example, the degree to which it involves employees in decision making. The means to socialize new
members are also much more effective. Apart from things like trainings and seminars, new members are
often told to follow a specific methodology. As they discover the merits of the method and start using it
independently, they become in fact socialized.
19
Note that this definition does not exclude that a company can have more than one culture: the sales department
might develop a (partially) different culture than the production department.
20
Note that even what seems to be the purest of conventions, which side of the road to drive on, has structural
origins. For an extensive study of this ‘convention,’ see Kincaid (1986).
23
B
A Note on ‘Differing Beliefs’ in Economic Modeling
The model in this paper differs in one respect from many economic models: the agents knowingly entertain differing beliefs (without having private information). The reason for this assumption is pragmatic:
differences in beliefs are at the heart of the issues studied here, and assuming common knowledge of differing beliefs is the most transparent and parsimonious way to study this question. Differing beliefs do
not contradict the economic paradigm: while rational agents should use Bayes’ rule to update their prior
with new information, nothing is said about those priors themselves, which are primitives of the model. In
particular, absent any relevant information agents have no rational basis to agree on a prior.21 Harsanyi
(1968), for example, observed that ‘by the very nature of subjective probabilities, even if two individuals
have exactly the same information and are at exactly the same high level of intelligence, they may very well
assign different subjective probabilities to the very same events’. The best argument for the traditional use
of common priors is Aumann’s (1987) argument that they allow us to ‘zero in on purely informational issues’.
Conversely, differing priors allow us to zero in on the implications of open disagreement and differing beliefs.
Note, finally, that the existence of the winner’s curse, any disagreement on the correct action when utilities
are aligned, and nearly all evidence on bounded rationality and biases in decision making imply empirically
the existence of differing priors. For a further discussion, see Morris (1995), the discussion between Gul
(1998) and Aumann (1998), Yildiz (2000), or Van den Steen (2002c).
The model with common priors It is actually theoretically possible to develop this model under
the common prior assumption. In particular, assume that:
• Agents have a common prior.
• All agents get a private signal about each action.
• All employees of a firm observe the manager’s signal.
• Agents cannot in any way observe each other’s beliefs.
This setup makes the agents behave as if they have differing priors. While it sticks pro forma to the tradition
of a common prior, it does so artificially. I therefore prefer to call a cat a cat and work with differing priors.
21
A traditional argument against differing priors is that rational people would argue until they reach agreement.
There are 3 reasons why this argument does not hold. First of all, any fact that is presented in the discussion brings
with it potentially differing priors about the relevance of the fact or the precision of the data. This leads immediately
to an infinite regress of arguments. Second, what matters for economic analysis is not so much whether agents can
theoretically come to agreement but whether they do at the time they have to take an action. Finally, as argued in
the main paper, much information is difficult, if not impossible, to communicate. Note also that under the common
prior assumption, agents with aligned utility functions should never disagree on the optimal actions to undertake.
On a sinking ship or in the heat of a battle (where survival is supposedly the common objective), there should always
be perfect agreement on the optimal course of action. This seems to contradict our intuition.
24
References
Aghion, P.,
and J. Tirole (1997):
“Formal and real authority in organizations,” Journal of Political Economy, 105(1), 1– 29.
Aumann, R. J. (1987): “Correlated Equilibrium as an Expression of Bayesian Rationality,” Econometrica, 55, 1– 18.
Aumann, R. J. (1998): “Common Priors : A Reply to Gul,” Econometrica, 66(4), 929– 938.
Baker, E. L. (1980): “Managing Organizational Culture,” Management Review, 69(7), 8– 13.
Baker, G. P. (1992): “Incentive contracts and performance measurement,” Journal of Political Economy, 100(3), 598– 614.
and
Baron, J. N., M. D. Burton,
M. T. Hannan (1999): “Engineering Bureaucracy: The Genesis of Formal Policies, Positions, and Structures in
High-Technology Firms,” Journal of Law, Economics, and Organization, 15(1), 1– 41.
and A. Schoar (2001): “Managing with Style: The Effect of Managers on Firm Policies,” Working paper.
and G. M. Stalker (1961): The Management of Innovation. Londen : Tavistock.
Burt, R. S., S. M. Gabbay, G. Holt, and P. Moran (1994): “Contingent Organization as a Network Theory: The Culture Performance Contingency
Bertrand, M.,
Burns, T.,
Function,” Acta Sociologica, 37, 345– 370.
and D. Gromb (1999): “On the Strength of Corporate Cultures,” European Economic Review, 43, 1021– 1037.
and J. I. Porras (1994): Built to last : Successful Habits of Visionary Companies. Harper & Row, New York.
Crawford, V., and J. Sobel (1982): “Strategic Information Transmission,” Econometrica, 50, 1431– 1452.
Carrillo, J. D.,
Collins, J. C.,
Crémer, J. (1993): “Corporate Culture and Shared Knowledge,” Industrial and Corporate Change, 3(3), 351– 386.
Deal, T.,
and A. A. Kennedy (1982):
Corporate Cultures: The Rites and Rituals of Corporate Life. Addison Wesley, Reading MA.
Dessein, W. (2001): “Authority and Communication in Organizations,” Forthcoming in the Review of Economics Studies.
and J. W. Lorsch (1983): Decision Making at the Top : The Shaping of Strategic Direction. Basic Books, New York.
and D. M. Jones (1974): “A Dynamic Allocation Index for the Sequential Design of Experiments,” in Progress in Statistics, ed. by
J. Gani, et al., and et. al. North-Holland, Amsterdam.
Donaldson, G.,
Gittins, J. C.,
Greif, A. (1994): “Cultural beliefs and the organization of society : A historical and theoretical reflection on collectivist and individualist societies,”
Journal of Political Economy, 102(5).
Gul, F. (1998): “A Comment on Aumann’s Bayesian View,” Econometrica, 66(4), 923– 927.
Harsanyi, J. C. (1968): “Games with Incomplete Information Played by ‘Bayesian’ Players, I-III, Part III. The Basic Probability Distribution of
the Game,” Management Science, 14(7), 486– 502.
Hermalin, B. E. (2001): “Economics and Corporate Culture,” in The International Handbook of Organizational Culture and Climate, ed. by C. L.
Cooper, S. Cartwright,
and P. C. Earley. John Wiley & Sons, New York, Working paper, University of California-Berkeley.
Kincaid, P. (1986): The Rule of the Road: An International Guide to History and Practice. Greenwood Press, Westport, CT.
Kotter, J. P.,
and J. L. Heskett (1992):
Corporate Culture and Performance. Free Press, New York.
Kreps, D. M. (1990): “Corporate Culture and Economic Theory,” pp. 90– 143.
Lazear, E. P. (1995): “Corporate Culture and Diffusion of Values,” in Trends in Business Organization : Do Participation and Cooperation Increase
Competitiveness, ed. by H. Siebert, pp. 89– 133. J.C.B. Mohr/Paul Siebeck, Tubingen.
Lee, D.,
and E. Van den Steen (2003):
“Best Practice or Worst Failure? On the Design of Knowledge Systems,” Working paper.
Martin, J. (1992): Cultures in Organizations: Three Perspectives. Oxford University Press, New York.
McCaskey, M. B. (1997): “Framework for Analyzing Workgroups,” HBS Case 9-480-009.
McGahan, A. M. (1999): “The Performance of US Corporations: 1981-1994,” The Journal of Industrial Economics, 47(4), 373– 398.
Morris, S. (1995): “The Common Prior Assumption in Economic Theory,” Economics and Philosophy, 11, 227– 253.
Mueller, D. C. (1990): “The Persistence of Profits in the United States,” in The Dynamics of Company Profits: An International Comparison, ed.
by D. C. Mueller. Cambridge University Press, Cambridge.
and R. H. Waterman (1982): In Search of Excellence : Lessons from America’s Best-run Companies. Harper & Row, New York.
and M. Visscher (1980): “Organization Capital,” Journal of Political Economy, 88(3), 446– 461.
Rob, R., and P. B. Zemsky (2002): “Social Capital, Corporate Culture and Incentive Intensity,” Rand Journal of Economics, 33(2), 243– 257.
Peters, T. J.,
Prescott, E. C.,
Rothschild, M. (1974): “A Two-Armed Bandit Theory of Market Pricing,” Journal of Economic Theory, 9, 185– 202.
Schein, E. H. (1985): Organizational Culture and Leadership. Jossey-Bass Publishers, San Francisco, CA.
Schwartz, H.,
and S. Davis (1981):
“Matching Corporate Culture and Business Strategy,” Organizational Dynamics, pp. 30– 48.
Sørensen, J. B. (2002): “The Strength of Corporate Culture and the Reliability of Firm Performance,” Administrative Science Quarterly, Forthcoming.
Van den Steen, E. J. (2002a): “Birds of a Feather ...,” Working paper, MIT-Sloan.
(2002b): “Costs and Benefits of Homogeneity,” Working paper.
(2002c): “Notes on Modelling with Differing or Heterogeneous Priors,” Working Paper, MIT-Sloan.
Van den Steen, E. J. (2002d): “Organizational Beliefs and Managerial Vision,” Working paper, MIT-Sloan.
Watson, Jr., T. J. (1963): A Business and Its Beliefs: The Ideas That Helped Build IBM. McGraw-Hill, New York.
and
Weber, R. A.,
C. F. Camerer (2001): “Cultural Conflict and Merger Failure: An Experimental Approach,” USC Center for Law, Economics,
and Organization Research Paper C01-10.
Yildiz, M. (2000): “Essays on Sequential Bargaining,” PhD Dissertation, Stanford.
25